Seminars: Dept Seminar

Issues in the Application of Latent Class Models for Survey Evaluation: An Agenda for Future Research

Abstract:

Latent class analysis is a useful tool for evaluating measurement error in surveys.  Under an assumed model, it provides estimates of classification error and measurement bias without requiring gold standard measurements or direct estimates of the true parameter values.  However, appropriate use of the methodology can be elusive due to problems with the underlying model assumptions and other estimation issues like unidentifiability, data sparseness, boundary estimates, and latent class "flippage."  This presentation will illustrate some uses of LCA for survey evaluation, discuss some of the issues associated with its use and outline an agenda for future research for statisticians who want to contribute to this important area measurement error research.